AI Solutions
The technology is chosen last.
There are five ways an AI outcome actually gets delivered. They are not tiers, and the fifth is not better than the first — they answer different shapes of problem, and the expensive mistake is picking one before you know which shape you have. All five are governed AI — a category, not a product. This page describes them plainly, including the ones we would talk you out of.
The boundary
We are Microsoft-centric about your environment and deliberately open about what we put to work inside it.
The tenant, identity, devices, data and permissions are Microsoft, and that is what we harden. What gets applied on top of that foundation is chosen against the problem. Sometimes that is Copilot. Often it is not.
The five options
One outcome, delivered with governed AI.
Governed AI is the category. The option inside it is chosen against the shape of the problem, never against a vendor relationship. They are not tiers, and the fifth is not better than the first.
1Copilot
For work that lives inside Microsoft 365 — documents, mail, meetings, chat. Not the answer when the knowledge it needs sits somewhere else.
2Company AI Accounts
For when your people are already using AI on personal accounts. Work done in a personal account leaves with the person who did it.
3Connected Assistants
For when something has to happen inside a system you already run, rather than simply be known.
4Workflow Platforms
For a process with steps, handoffs and exceptions. Two routes, and the right one depends on where the work lives.
5Tailored Builds
For when nothing off the shelf matches how the business actually delivers. The slowest and most expensive of the five.
Which one is yours
Four questions that usually settle it.
The quickest test
One question splits three ways. If asking is enough you are at one or two. If something has to happen inside a system you are at three or five. If the problem is a process with steps you are at four.
The quickest test
What does the work need?
- Asking is enough1 or 2A person reads the answer and acts on it themselves
- Something has to happen3 or 5An action lands inside a system you already own
- It is a sequence4Steps, a handoff, and an exception somebody judges
The four questions below work the same way. None of them is about the technology, and none of them needs you to know what the five options are.
Where does the work live?
If the answer is mail, documents and meetings, start at option one. If it is a line-of-business system, look at three. If it is a process that crosses several, look at four.
Is the problem thinking, or is it a sequence?
Thinking work points at one or two. A sequence with handoffs and exceptions points at four.
Does something have to happen, or just be known?
If the outcome is an action inside a system, you are at three or five. If it is an answer, you may be at one, two or four.
What are your people already doing?
This is the most useful question and the least often asked. If several people independently reached for the same tool for the same task, they have already told you where the demand is — and that finding usually names the option for you.
Two, three and five can look alike. They are not.
All five are governed AI. All three of these can use the same underlying model. What differs is how much is built around it. Two builds nothing — the model reads and reasons, hands a person an answer, and the person acts on it. Three builds connections — the same model, given permission to act inside systems you already own. Five builds an application — for when the thing you need has no home in any product you can buy.
The quickest test. Can a person get what they need by asking, and then doing it themselves? That is two. Does something have to happen inside a system you already run? That is three. Does the work have nowhere to live at all? That is five.
None of these is answerable from a brochure, which is why we do not ask you to answer them alone.
If you have no preference yet
If you do not know yet, you are in the normal position.
Most people arrive here without a preference. That is not a gap in your homework — it is what an assessment is for, and the sequence is the same for everyone.
First we establish what you already own and pay for, how identity, devices and permissions are configured today — read-only, against your own tenant, and what AI is already running in your organization whether or not anyone approved it. Then we fix what the evidence says has to be fixed. The option gets chosen after that, and by then it usually chooses itself.
- 1CopilotWork that lives inside Microsoft 365
- 2Company AI AccountsReasoning over material Copilot does not reach
- 3Connected AssistantsActing inside the systems where work happensAssumes one or two already exists
- 4Workflow PlatformsA process with steps, handoffs and exceptionsAssumes one or two already exists
- 5Tailored BuildsNothing off the shelf matches how you deliverAfter you know what you cannot buy
The hardened tenant — under all five, and not one of them
Same row, same height, on purpose. The order is the shape of the problem, not a ranking — the fifth is not better than the first, and most organizations do one thing and stop there.
They are not tiers. But some of them do rest on others.
- The hardened tenant sits under all five.
- It is not an option on the list. It is the floor, and nothing intelligent should run on an environment nobody has secured.
- One and two put capability in people’s hands.
- They are the fastest to stand up, because neither of them requires anything to be built first.
- Three and four assume one or two already exists.
- You cannot responsibly connect a model to your finance system before deciding which model, under what terms, and with what access.
- Five comes last on purpose.
- The right time to build something is after you know what you cannot buy.
Most organizations do one thing, measure it, and then decide whether there is a second. Stopping after the first is a legitimate place to stop, not a failure to progress.
What this means commercially
For the technology partner
Only one of the five is a product a competitor can quote line for line.
If Copilot is the whole offer, you are selling one SKU at Microsoft’s margin on Microsoft’s roadmap, and the provider down the street is selling the identical thing at the identical price. Governed AI deployment, integration work, platform configuration and tailored builds are not comparable in that way — and work that is not comparable is priced on judgment rather than on the going rate.
You cannot win a price comparison. You can only stop being in one.
For the business
The option changes the scope and the price. It does not change the shape of the work.
Whichever of the five is chosen, the engagement looks the same: one outcome, a baseline measured before anything is switched on, a named owner, and the result published. That is the part we hold constant, because it is the part that determines whether anything actually improved.
Where this sits
The option is chosen in step four. Not before it.
Everything on this page belongs to the second half of the method. The first half decides whether any of it is safe to switch on, and which outcome is worth doing first.
Foundation gatesPassed once, not optional
- 01UnderstandThe assessment
- 02SecureCyber hardening
- 03PrepareAI readiness and governance
Progress cycleRepeats, for every outcome
- 04ApplyAI Solutions — one outcome
- 05OrchestrateAI Solutions — connected
- 06EvolveAI Solutions — measured, then improved
This is why the technology is chosen last. By the time an option is named, the environment is understood, the foundation is hardened, and one outcome has an owner and a baseline.
Where to start
The assessment is how you choose.
It establishes what you already own and pay for, how identity, devices and permissions are configured today — read-only, against your own tenant, what AI is already in use and by whom, and which single outcome is worth doing first. The option follows from that. It does not lead it.
Technology partners begin with the MSP Secure AI Practice Assessment. Businesses begin with the Secure AI Operations Assessment, offered through your technology partner where one is in place.